SkyNet: A Deep Learning Architecture for Intra-hour Multimodal Solar Forecasting with Ground-based Sky Images

Ruan, Guoping, Chen, Xiaoyang, Li, Yiheng, Lim, Eng Gee, Fang, Lurui, Jiang, Lin ORCID: 0000-0001-6531-2791, Du, Yang and Wang, Fei (2026) SkyNet: A Deep Learning Architecture for Intra-hour Multimodal Solar Forecasting with Ground-based Sky Images RENEWABLE ENERGY, 256. 124354-. ISSN 0960-1481, 1879-0682

Text Skynet.pdf - Author Accepted Manuscript Available under License Creative Commons Attribution. Download (8MB) | Preview

Abstract

The increasing penetration of photovoltaic systems introduces critical challenges to grid transient stability, primarily due to rapid power fluctuations induced by localized cloud dynamics. While intra-hour solar forecasting using ground-based sky images has emerged as a pivotal approach for mitigation strategy, it remains fundamentally constrained in addressing three crucial limitations: (1) low capability of detecting cloud dynamics for time-series forecasting, (2) probabilistic uncertainty quantification essential for risk-aware grid management, and (3) spatially resolved spatial forecasting critical for distributed energy resource coordination. We propose SkyNet, a unified multimodal deep learning framework that integrates time-series, probabilistic, and spatial forecasting within a single model. To capture local details and long-range dependencies while enabling efficient multimodal feature fusion, the Dilated Attention With Neighborhood module was proposed. Meanwhile, a unified loss function was designed to jointly train all tasks. Experimental results demonstrate that SkyNet delivers competitive or superior accuracy across horizons compared with the state-of-the-art benchmark models, offering an efficient and comprehensive forecasting solution for high-renewable power systems.

Item Type:	Article
Uncontrolled Keywords:	Photovoltaics, Solar forecasting, Sky images, Multimodal forecasting
Divisions:	Faculty of Science & Engineering Faculty of Science & Engineering > School of Engineering Faculty of Science & Engineering > School of Engineering > Electrical Engineering and Electronics
Depositing User:	Symplectic Admin
Date Deposited:	12 Feb 2026 15:38
Last Modified:	16 Jun 2026 20:24
DOI:	10.1016/j.renene.2025.124354
Related Websites:	Author Publisher
URI:	https://livrepository.liverpool.ac.uk/id/eprint/3197010
Disclaimer:	The University of Liverpool is not responsible for content contained on other websites from links within repository metadata. Please contact us if you notice anything that appears incorrect or inappropriate.